GSWABE: faster GPU‐accelerated sequence alignment with optimal alignment retrieval for short DNA sequences

Abstract

In this paper, we present GSWABE, a graphics processing unit (GPU)-accelerated pairwise sequence alignment algorithm for a collection of short DNA sequences. This algorithm supports all-to-all pairwise global, semi-global and local alignment, and retrieves optimal alignments on Compute Unified Device Architecture (CUDA)-enabled GPUs. All of the three alignment types are based on dynamic programming and share almost the same computational pattern. Thus, we have investigated a general tile-based approach to facilitating fast alignment by deeply exploring the powerful compute capability of CUDA-enabled GPUs. The performance of GSWABE has been evaluated on a Kepler-based Tesla K40 GPU using a variety of short DNA sequence datasets. The results show that our algorithm can yield a performance of up to 59.1 billions cell updates per second (GCUPS), 58.5 GCUPS and 50.3 GCUPS for global, semi-global and local alignment, respectively. Furthermore, on the same system GSWABE runs up to 156.0 times faster than the Streaming SIMD Extensions (SSE)-based SSW library and up to 102.4 times faster than the CUDA-based MSA-CUDA (the first stage) in terms of local alignment. Compared with the CUDA-based gpu-pairAlign, GSWABE demonstrates stable and consistent speedups with a maximum speedup of 11.2, 10.7, and 10.6 for global, semi-global, and local alignment, respectively. GSWABE: GPU-ACCELERATED DNA SEQUENCE ALIGNMENT 959 especially for large-scale datasets. This has therefore driven a substantial amount of research to parallelize pairwise alignment on high-performance computing architectures ranging from looselycoupled to tightly-coupled ones, including clouds [12], clusters [13,14], and accelerators [15,16]. Among these architectures, accelerators, including single instruction multiple data (SIMD) vector processing units (VPUs) affiliated to CPUs, field programmable gate arrays (FPGAs), and general-purpose GPUs, have recently been the predominant techniques.The SIMD VPUs affiliated to CPUs are the most widely used techniques. Two general approaches have been investigated to meet the computational features of SIMD vectors: one is the inter-task (or inter-sequence) parallelization model and the other is the intra-task (or intra-sequence) model. The inter-task model performs multiple alignments in individual SIMD vectors with one vector lane computing one alignment (e.g. [17]). The intra-task model computes in parallel the alignment of a single sequence pair in vectors based on two computational patterns: vectorized computation parallel to minor diagonals in the alignment matrix [18] and vectorized computation parallel to the query sequence in a sequential [19] or striped [20] layout. The two kinds of models provide a general framework for other accelerators with SIMD VPUs, including Cell Broadband Engine and general-purpose GPUs. Few implementations [21,22] have been proposed on Cell Broadband Engine, and all of them are based on the intra-task model with the striped layout. On generalpurpose GPUs, open graphics library was initially...

HPMA: High-performance metagenomic alignment tool, on a large-scale GPU cluster

“…We also compiled a test set to allow us to compare the per formance of HPMA with the published results for GSWABE [7]. We generated three more datasets which have similar attributes to those used to evaluate GSWABE.…”

“…The motivation for selecting these two types of test sets was to investigate the performance of HPMA in the metagenomic contexts of very similar sequences and very dissimilar sequences. The third test set was employed to allow a comparison with published results for GSWABE [7]. The datasets in this test set are derived from the same Escherichia coli K12 MG1655 genome (GenBank accession number NC_000913) using comparable sampling parameters as that for GSWABE.…”

“…The methodology section describes the data sets and the design and implementation of the HPMA kernel. This is followed by a section on perfor mance evaluation which includes a comparison of HPMA with GSWABE, a competing GPU alignment tool [7]. Finally, we summarize the results presented in the paper and place HPMA in the context of a larger metagenome gene identification framework.…”

Accelerating Exhaustive Pairwise Metagenomic Comparisons

“…mpi-BLAST [3] for distributed memory or TERABLAST [4] for its use on FPGAs). Parallelization of sequence alignment algorithms have also been applied to GPUs such as GSWABE [5] or CUSHAW2-GPU [6]. FPGAs have also been employed to accelerate sequence comparisons (e.g.…”

Optimized and Portable FPGA-Based Systolic Cell Architecture for Smith-Waterman-Based DNA Sequence Alignment

“…The mechanisms and specifics of the alignment of short DNA sequences are sometimes divergent with the goals of the alignment of large sequences. The authors in [10] developed a platform that through GPU could accelerate the alignment of short sequences. Other architectures are also explored for accelerating the alignment of DNA sequences.…”